A Comparison of Digital Transmission Techniques Under Multichannel Conditions at 2.4 GHz in the ISM BAND. Fabien_Mulot (ONERA, TESA/SUPAERO, [email protected]) Vincent Calmettes (SUPAERO, [email protected])

ABSTRACT In order to meet the observation quality criteria of the micro-UAVs, and particularly in the context of the « Trophée Micro-Drones », SUPAERO is studying technical solutions to transmit a high data rate from a video payload onboard a micro-UAV. The laboratory has to consider the impact of multipath and shadowing effects on the emitted signal (fig1.) and therefore it has to select fading resistant transmission techniques. The following of this paper discusses the study.

a. Reflection b. Shadowing c. Line of Sight

2. CARACTERIZATION OF THE VIDEO STREAM 2.1 Description of the payload

The 640x480 pixels images are coded on 8 bits (256 grey levels). Then they are compressed in a JPEG format. See [Bur,03] for an example of a previously developed payload. JPEG is a lossy compression method because of the use of a quantification matrix. The losses are characterized by the creation of blocs on the image.

a

b

The study had to reveal an optimum trade-off between three parameters, namely: the characteristics of the video stream, the complexity of the modulation and coding scheme, and the efficiency of the transmission, in term of BER.

.

c

Figure 1. The wireless propagation landscape

1. INTRODUCTION First, with the objective of achieving a video stream size around 1.5 Mbits.s-1 without coding, we defined a number of acceptable video characteristics in term of: refreshing rate, image resolution, and compression technique complexity. Then the mobile propagation channel has been characterized. We evaluated under Matlab/SIMULINK different transmission schemes (OFDM, Spread spectrum with rake receiver, QPSK with equalization) and different channel coding techniques (convolutional codes, ReedSolomon codes).

2.2 The different envisaged compression rates

The system will allow to switch between two modes. The first one, used for example in approach phases, will consist in a low resolution and a high image refreshing rate. The second one, used for more detailed scenes, will consist in a high resolution and a lower refreshing rate. Different compression rates have been selected (table1). A 10% extra margin has been added because the compression efficiency depends on the nature of the image. An image with large surfaces of the same color (low entropy) will give a smaller file size when compressed than an image with more colors (higher entropy) for the same quantification matrix. Image Uncompressed A B C D

Compression Rate 1 33.5 20 10 8

Size [Ko] 302 9 15 28 38

+10% margin [Ko] 10 16.5 31 42

Table 1. Image Size Vs Compression Rate

1

Image A B C D

Bit rate at 14 i/s 1.12 Mbits/s 1.848 Mbits/s 3.472 Mbits/s 4.704 Mbits/s

Bit rate at 2 i/s 160 Kbits/s 264 Kbits/s 496 Kbits/s 672 Kbits/s

D

C

B

Table 2. Bit rate Vs Image Rate

The table 2 shows different bit rates as a function of the number of images per second. These values will be used later in this paper.

***

The first picture below represents the uncompressed image. The pictures A to D correspond to the selected compression rates in tables 1 and 2. They show the impact of compression losses on the ability to read a text (Electronic) and to distinguish a face or a shape in a portion of the picture. The 33.5% compression rate highly the picture [A]. The face and the inscription can’t be distinguished. picture still contains enough details the micro drone to navigate.

degrades smallest But the to allow

Unsurprisingly, the other compression rates give an increased quality but the data rates are bigger too. They can be used to provide more detailed scenes depending on the resolution required. But the counterpart will be a lower image refreshing rate.

D

Picture 1. Visual effects of different compression rates

The strategy will be to provide a 1.12 Mbits/s video bit rate and therefore, either: -

14 i/s in a low resolution [table1 A] or 3.3 i/s in a high resolution [table1 D].

3. MODELISATION OF THE WIRELESS CHANNEL 3.1 Transmission context The band used for data transmission is the industrial, scientific, and medical ISM band. The usable bandwidth is around 79 MHz between 2.4 GHz and 2.485 GHz. This band is license free and standards using this band like Bluetooth and Wifi are open. Concerning the emitted power, the maximum EIRP authorized by the ANFR (Agence Nationale des Frequences) is 10mW outdoor and 25 mW indoor. 3.2 General channel model One subdivide the multiplicative fading processes in the channel into three types of fading: Path Loss, Shadowing and Multipath Fading. The Additive White Gaussian Noise is then taken into account.

A

Path Loss

Shadowing

Multipath Fading

AWGN

2

Figure 2. Wireless communication channel model

(4πd )

Path loss is given by L= λ

2

with the same delay. Each tap represents a single beam. The gains αn are varying in time independently of each other, according to the standard following laws: Rayleigh distribution, (Eq 2), for Non Line of Sight path and Rice distribution, (Eq 3), for LOS conditions. The worst case is the non LOS transmission where less power is available.

(Eq1)

d distance from the emitter to the receiver. Shadowing changes more rapidly than path loss, with significant variations over distances of hundreds of meters and generally involving variations up to around 20 dB. The PDF of the attenuation process is log-normal; that is, the attenuation measured in decibel has a normal distribution.

PR (α ) = (α / σ 2 )e − (α pR (α ) = (α / σ )e 2

2

/( 2σ 2 ))

(2)

− (α 2 + s 2 ) / 2σ 2

I 0 (α s / σ ) (3) 2

Table 3 gives a typical power-delay profile for mobiles communications. The first coefficient of the filter follows a Rice or a Rayleigh distribution for respectively a LOS or a non LOS transmission. In any cases, the other coefficients follow a Rayleigh law.

Multipath fading involves faster variations of a scale of a half-wavelength (6.25 cm at 2.4 GHz) and generally introduces variations as large as 35 to 40 dB. It results from the constructive and destructive interference between multiple waves reaching the receiver.

Tap Delay Average µs Power dB 1 0.0 0 2 3 4 5 6

0.31 0.71 1.09 1.73 2.51

-5 -9 -11 -15 -20

Law

PSD

Rice or Rayleigh Rayleigh Rayleigh Rayleigh Rayleigh Rayleigh

Jakes Jakes Jakes Jakes Jakes Jakes

Table 3. Standard channel profile for UMTS Figure 3. The three scales of mobile signal variation

3.2. Countering the Shadowing Effects 3.3. Multipath Channel Model The figure 4 shows the standard form of the multipath channel model. X(t)

τ1 α1(t)

τn

τ2

⊗

α2(t)

⊗

αn(t)

⊗

Σ

Y(t)

Figure 4. Standard Channel Model

The effects of scatterers in discrete delay ranges are lumped together into individual taps

3.3.1. Second order fading statistics In order to model the temporal autocorrelation of fades (depending on the speed of the mobile v), the following assumptions are made. The antenna of the receiver is omnidirectional. The arrival angle of waves is uniformly distributed around the receiver. Therefore, a suitable law is the classical Jakes’ power density spectrum (Eq 4).

S(f)= E0 4πf m with

1 2

1−(f / fm) v f = fc cos(θ ) c

f < fm

(4)

(5)

fm maximum Doppler shift Eo energy constant

3

3.3.2. Rice and Rayleigh fading models The rician PDF is given by:

pr (α ) =

k=

S

2kα

2

2σ

2=

S

2

µ

e(

− kα S

2

2

) e( − k ) I 0 (

2

2σ

2

and

2kα ) S

2

(6)

2

2

E r(t) = P =2σ + µ (7)

S Magnitude of the LOS component 2 Variance of either the real or imaginary component. P average power given in table1 While µ=0 corresponds to the Rayleigh PDF, Rice and Rayleigh complex processes models are the following: Complex Gaussian Noise

S(f)

(t)

Figure 5.a Rice complex process

S(f)

Tc = 9 /(16Π fm)

(8)

The speed of the Micro UAV is in the range 0 - 50 km.h-1. Eq (5) gives the corresponding Doppler fm ranging from 0 to 110Hz. Assuming a 110 Hz maximum Doppler , EQ (8) implicates Tc equal to 1.6 ms. Therefore, signals with a rate lower than 625 Symb.s-1will propagate in a fast varying channel; the channel varies during the propagation of the symbol and Doppler spread occurs. Signals with a higher rate will propagate in a slow varying channel without Doppler spread. b) Coherence bandwidth

Jakes’ PSD

Complex Gaussian Noise

When the response of a channel is timevariant, Doppler spread occurs. Signals which have less than Tc are received approximately undistorted by Doppler spread. [Saunders, 99] gives an approximation of the coherence time for the classical channel.

(t)

Jakes’ PSD

Figure 5.b Rayleigh complex process

The figure 6 shows the mobile channel model obtained using a 6 rays urban model for GSM

A closely related parameter to the coherence bandwidth Bc of the fading channel is the multipath spread Tm. Therefore the channel will be considered as non frequency selective if T>>Tm, where T is the symbol duration. In this case the channel is said to be flat in the frequency domain. [Jakes,94] shows that assuming a classical Doppler spectrum for all components, the coherence bandwidth is:

Bc =

3 22Π τ rms

τ rms =

τ0 = Figure 6. 6 rays urban model for GSM

3.3.3. Mobile radio channel characteristics at 2.4Ghz

1 PT

1 PT n i =1

n i =1

(9)

Piτ i − τ 0 2

Piτ i and

(10)

PT =

n i =1

Pi

(11)

τrms root mean square delay spread. τ0 mean delay.

a) Coherence time of the channel

4

The table 4 shows the values obtained for a set of areas corresponding to the competition context. The bandwidth needed for the video transmission is wider than the values of Bc shown in table 4. A frequency selective behavior of the channel must be expected. Model GSM 12 rays GSM 6 rays UMTS 6 rays Channel A UMTS 6 rays Channel B

Area Urban Area Urban Area Macro cell Low delay spread Macro cell High delay spread

Bc 25 034 Hz 23 034 Hz 67 600 Hz 6 230 Hz

Table 4 Coherence Bandwidth for different standard mobile channel models

Figure 7.a shows the channel model for a 110 Hz maximum Doppler. Figures 7.b and 7.c show a realization of the channel respectively at a given instant and for a given frequency. For video transmission, it is reasonable to envisage an occupied bandwidth between 2Mz an 70Mhz, depending on the transmission technique. In that case, figure 7.b confirms the frequency selective behavior of the channel. The emitted video signal undergoes several deep fades up to 20 dB.

Symbol duration 1µs

Symbol undergoing a fade

Figures 7.c Time domain

Figure 7.c shows that deep fades are also experienced in the time domain and that groups of symbols can be totally lost. Thus, fade mitigation techniques like time diversity must implemented in order to use the information given by the different delayed signals, figure 7.d. (e.g Rake Receiver).

Figure 7.d Multiple superimposed delayed signals 3.3.4. Validity the power delay profiles

Figure 7.a 12 rays GSM channel model for urban area

Deep Fades

25 kHz Coherence Bandwidth

Example of a 4 MHz occupied bandwidth for video transmission

Figures 7.b Frequency domain

The simulation campaigns use standard power delay profiles for semi urban areas taken from studies made for the GSM and the UMTS standards. Concerning the speed of the mobile, these models are made for a 250 Km.h-1 maximum mobile speed. Consequently, these profiles are suited to the micro UAV speed of 50 Km.h-1. Nevertheless, these profiles must be taken cautiously. The GSM and UMTS standards use a lower frequency band, respectively 900-1800 MHZ and 1920-2170 Mhz and they provide a lower bit rate than the transmission payload of the micro UAV. The UMTS profiles are the most representative and they give a good idea of the length of impulse response of the transmission channel, around 5 µs, for a first modelisation of the system under Simulink. To be more precise further measures should be done under field conditions at 2.4 GHz. 3.3.5. Classification of the channel

5

The developments made previously allow to classify the propagation channel as slow fading – frequency selective. [Figure 8] Bs

Bc

Frequency Selective-Fast fading

Frequency Selective-Slow fading Flat-Slow fading

Flat-Fast fading Bd=1/Tc

Bs

Transmitted Baseband Signal Bandwidth

Figure 8. Matrix illustration of the different types of fading

4. Channel coding strategy Channel coding stages are composed of a cyclic (204/188) Reed Solomon coder followed by an first (outer) interleaver, a [171,133] convolutional coder of constraint length 7 and rate ½, and a second (inner interleaver). 4.2. Inner interleaving.

In a multipath fading environment errors occurs by burst. Block codes and convolutional codes are effective over memoryless channels where errors are random. Using an interleaver the channel can be made memoryless and FEC schemes can be used.

3.4. Link Budget

4.3. Convolutive coding.

The following table synthesizes the link budget. It assumes a QPSK mapping, the use of training sequences occupying 1/6 of the time slots, and the channel coding described in 4.

Convolutive codes are used for their good errors correcting properties. The generator matrices of convolutive codes must be non catastrophic because these matrices can generate an infinite number of errors. A polynomial generator matrix G(x) is non catastrophic if and only if ∆k(x)=xS, S 0, where ∆k(x) is the greatest common divisor of all determinants of all kxk submatrices of G(x), and where k is the number of inputs of the encoder. [Bos, 99]

3.5 Required Eb/N0 (See 4.8) 300K T0 ambiant temperature 457K Te receiver equivalent noise temperature 0dB Ge emitter gain 12dB Gr receiver gain 2.4Ghz Fp 1Km D emitter-receiver distance 1.02 Msb/s Rs symbol rate Pr min = Rs.k.No(Te+T0).(Eb/N0) -102 dBm -13.95dBm For a LOS transmission 2 Pe = Pr.(1/Ge).(1/Gr).( (4.π.D)/ λ) 6 dBm For a Non LOS transmission 4 mW (Rayleigh fading). 4 Pe ~ Pr.(1/Ge).(1/Gr).(D) 11 to For a Non LOS transmission (Rayleigh fading) plus 5 to 20dB of 26dBm shadowing.

It appears that the required power must considerably increased in the case of a Non LOS transmission, and even more for when shadowing occurs. A solution could be to use a variable gain in the emitter to increase the power in the case of a strongly attenuated transmission.

4.4. Puncturing

A punctured convolutional code is one in which some bits are discarded at the transmitter to reduce the amount of data to be transmitted. Convolutive punctured codes allow a dynamic adaptation of the bit rate. 4.5. Interleaving

The interleaver reduces the size of large bursts of errors by spreading them. It must be chosen long enough to spread large bursts of errors occurring, for example, during deep fades due to shadowing.

4.6. RS coding.

The (204/188) RS code corrects the residual bursts of errors. The RS symbols are composed

6

of 8 bits. A (204,188) RS code corrects t=(204188+1)/2=16 symbol errors. The target BER after the RS decoder is 10-7 which means that the BER after the convolutive coder must be 10-3 maximum.

satisfies the Nyquist’s criterion and is widely used, in digital communications, to limit ISI. The Square Root Raised Cosine filter allows to split this filter between the transmitter and the receiver. The RC filter has also the property to reduce the occupied transmission bandwidth B. B=Rs.(1+ ) where Rs is the QPSK symbol rate. The Roll Off factor can be adjusted to meet the bandwidth requirements. 0